Transcript What To Do With 1,000,000 Quasars Gordon Richards Drexel University With thanks to Adam Myers (Illinois), Alex Gray, Ryan Reigel (Georgia Tech), Bob Nichol (Portsmouth),

What To Do With
1,000,000 Quasars
Gordon Richards
Drexel University
With thanks to Adam Myers (Illinois), Alex Gray, Ryan Reigel
(Georgia Tech), Bob Nichol (Portsmouth), Joe Hennawi
(Berkeley), Don Schneider (Penn State), Michael Strauss
(Princeton), Alex Szalay (JHU), and a host of other people from
the SDSS Collaboration
Want winds instead? See arXiv:astro-ph/0603827
Additional References
Binaries/Small Scales
Myers et al. 2007b, Clustering Analyses of 300,000 Photometrically
Classified Quasars. II. The Excess on Very Small Scales, ApJ, 658, 99
Myers et al. 2008, Quasar Clustering at 25 kpc from a Complete Sample of
Binaries, ApJ, submitted
Hennawi et al. 2006, Binary Quasars in the Sloan Digital Sky Survey:
Evidence for Excess Clustering on Small Scales, AJ, 131, 1
Comic Magnification
Scranton et al. 2005, Detection of Cosmic Magnification with the Sloan
Digital Sky Survey, ApJ, 633, 589
Integrated Sachs-Wolfe Effect
Giannantonio et al. 2005, High redshift detection of the integrated SachsWolfe effect, PhysRevD,74f3520
Giannantonio et al. 2008, Combined analysis of the integrated Sachs-Wolfe
effect and cosmological implications, arXiv0801.4380
99 year lease!
Name a
QUASAR for
the low, low
price of $99.99.
You get:
• Over 10 billion stars
• 1 supermassive
black hole
• loads of extras
The Sloan Digital Sky Survey
(SDSS) Quasar Sample
Spectra of ~100,000 quasars in 10,000 sq. deg.
i < 19.1 for z<3.0
i < 20.2 for z>3.0
Both color and radio selection
See Richards et al. 2002, AJ,
123, 2945 for details
So far: ~77,000 quasars from z=0 to z=6.4.
See Schneider et al. 2007, arXiv:0704.0806
Quasar
Surveys
Status
Hasinger et al. 2005
Optimizing Quasar Surveys
X-ray/IR surveys are deep enough (up to a few 1000
AGN/sq. deg.), but not wide enough.
Optical surveys are wide enough, but not deep enough.
SDSS
.
Need deeper optical surveys and/or larger area X-ray/IR surveys.
Can We Do Better?
Yes, we can.
Why We Need To
Do Better
Merger Scenario w/ Feedback
Merge gas-rich
galaxies, forming
buried quasars,
feedback expels
the gas, revealing
the quasar and
eventually shutting
down accretion.
Hopkins et al. 2005
How Can We Test This?
• The Quasar Luminosity Function
• active lifetime
• accretion rate
• MBH distribution
• Quasar Clustering
• L, z dependence
• small scales
Quasar Luminosity Function
Space
density of
quasars as a
function of
redshift and
luminosity
Typically fit
by double
power-law
Croom et al. 2004
Density Evolution
Number of
quasars is
changing as
a function of
time.
Luminosity Evolution
Space density
of quasars is
constant.
Brightness of
individual
quasars is
changing.
Hopkins et al. 2005
Most QLF models assume they are either “on” or “off” and
that there is a mass/luminosity heirarchy.
Hopkins et al.: quasar phase is episodic and “all quasars
are created equal” (with regard to mass/luminosity).
The SDSS QLF
SDSS, though
relatively
shallow, allows
us to determine
the QLF from
z=0 to z=5 with
a single dataset.
QLF slope
flattens at high-z.
Not PDE, PLE
Richards et al. (2006)
Understanding the High-z QLF
The change of the bright slope in the QLF at high redshift means
the distribution of intrinsic luminosities is broader at high redshift.
Hopkins et al. 2005
Richards et al. 2006
We Can Do Better
Hopkins, Richards, & Hernquist 2007
The Future: Efficient Target
Selection + Photo-z’s
Current selection techniques for quasars are
inefficient in the optical (~50-80% success rate).
It takes MUCH longer to take spectra than to get
photometry.
More efficient (~95%) selection algorithms coupled
with accurate photometric redshift techniques can
make spectroscopy nearly obsolete.
red
Traditional
Quasar
Selection
M
z < 2.2 quasars
F
z > 3 quasars
blue
A
blue
red
Spitzer-IRAC (Mid-IR) Selection
GAL
AGN
STAR
e.g. Lacy et al. 2004, Stern et al. 2005
How Can We Do Better?
Non-Parametric Bayesian Classification
with Kernel Density Estimation
(aka NBCKDE)
Richards et al. 2004, ApJS, Efficient
Photometric Selection of Quasars from the
SDSS: 100,000 Quasars from DR1, 155, 257
Given two training sets, Here quasars and stars
(non-quasars), and an unknown object, which
class is more likely?
“NBC”: Bayes’ (1763) Rule
P(x | Star)P(Star)
P(Star | x) 
P(x | Star)P(Star)  P(x | QSO)P(QSO)
Where
•
•
•
•
•
•
•
•
x = N-D colors
P(Star|x) = probability of being a star, given x
P(x|Star) = probability of x, drawing from stars training set
P(x|QSO) = probability of x, drawing from QSO training set
P(Star) = stellar prior
P(QSO) = quasar prior
P(Star) + P(QSO) = 1
Star if P(Star|x)>0.5, QSO if P(Star|x)<0.5
“KDE”: Kernel Density Estimation
N

1
PDF   K h x  x i
N i
 z 2 
K h (z)  exp  2 
2h 

x
xi
But Naïve KDE is O(N2)
Dual-tree Method
• Tree building is O(N log N); usually fast in comparison
to the rest of computation
• Classification of 500k objects in ~900 sec for
reasonable bandwidths
• See Gray, Riegel in Compstat 2006
Separating Quasars from Stars
DR6 Results
including high-z
840,000 – 1,060,000
quasars
Richards et al. 2008
Quasar Photo-z
z=1.5
u
g
r
i
z
Photometric Redshifts
Photometric
redshifts are
80% accurate
to within 0.3
Richards et al. 2001
Weinstein, Richards et al. 2004
SDSS vs. Johnson-Morgan/Kron-Cousins
SDSS+UKIDSS+IRAC
z=1.5
Hα plus slope change makes for robust photo-z
LSST
LSST corp.
QSO Detection With Time
1955-1990: Slow!
Methods include
radio/UVX detection
A. Myers
QSO Detection With Time
1990-2000: Multi-Fiber
Spectrographs/Plate
Scanning Machines
QSO Detection With Time
1995-2002: Long-term
Surveys 5 years yields
~80,000 QSOs
QSO Detection With Time
2002-2010: Million+
QSOs via Photometric
Classification?
Autocorrelation Function ()
• Red Points are, on
average, randomly
distributed, black points
are clustered
• Red points: ()=0
• Black points: ()>0
• Can vary as a function of,
e.g., angular distance, 
(blue circles)
• Red: ()=0 on all scales
• Black: () is larger on
smaller scales
A. Myers
•CDM P(k) projected
across redshift
distribution yields good
fit to shape of data.
• Linear bias (bQ=1)
ruled out at high
significance.
• Fitting for stellar
contamination improves
fit on scales larger than
a degree. Implied star
fraction ~< 5%
•For CDM cosmology,
quasar bias evolves as
a function of redshift
(Significance of
detection of evolution
>99.5% using only DR4
KDE data set).
• Detection in good
agreement with earlier
results from
independent
spectroscopic data (2dF
QSO redshift survey).
• Use ellipsoidal collapse
model (Sheth, Mo &
Tormen, 2001, MNRAS,
323, 1) to turn estimates of
bQ into mass of halos
hosting UVX quasars.
• Find very little evolution in
halo mass with redshift.
• Our mean halo mass of
~5x1012h-1MSolar is halfway
between characteristic
masses from Croom et al.
(2005, MNRAS, 356, 415)
and Porciani et al. (2004,
MNRAS, 355, 1010).
Hopkins+05 ApJ, 630, 716
“an observational probe that differentiates
quasars based on their host galaxy properties
such as … the dependence of clustering of
quasars on luminosity, can be used to
discriminate our picture from older models.”
Lidz et al. 2006
• Quasars accreting over a wide
range of luminosity must be
driven by a narrow range of
black hole masses
• M- relation mean a wide range
of quasar luminosities will then
occupy a narrow range of MDMH
Luminosity Evolution
bias
• Very little
dependence of
quasar clustering
on absolute
magnitude of the
quasar population
(Myers et al. 2007)
using large SDSS
photometric sample
Mg
Luminosity Evolution
• Similarly from the
SDSS+2dF=2SLAQ
quasar survey.
da Angela et al. 2008
What We (Used To) Expect
1. Galaxies (and their DM halos) grow through hierarchical
mergers
2. Quasars inhabit rarer high-density peaks
3. If quasars long lived, their BHs grow with cosmic time
4. MBH-σ relation implies that the most luminous quasars are
in the most massive halos.
5. More luminous quasars should be more strongly clustered
(b/c sample higher mass peaks).
6. QLF from wide range of e and narrow BH masses range
or wide range of BH masses (DMH masses) and narrow e
What We Get
1. Galaxies (and their DM halos) grow through hierarchical
mergers, but with “cosmic downsizing”
2. Quasars always turn on in potential wells of a certain size (at
earlier times these correspond to relatively higher density
peaks).
3. Quasars turn off on timescales shorter than hierarchical
merger times, are always seen in similar mass halos (on
average).
4. MBH-σ relation then implies that quasars trace similar mass
black holes (on average)
5. Thus little luminosity dependence to quasar clustering (L
depends on accretion rate more than mass).
6. Need a wide range of accretion efficiencies for a narrow
range of MBH to be consistent with QLF.
Conclusions
• Identification of large numbers of faint,
quasars is possible using novel statistical
methods
• Use of such methods will be crucial in
the LSST era
• The resulting samples are extremely
useful for testing the merger models of
quasars
• More to come!
Additional References
Binaries/Small Scales
Myers et al. 2007b, Clustering Analyses of 300,000 Photometrically
Classified Quasars. II. The Excess on Very Small Scales, ApJ, 658, 99
Myers et al. 2008, Quasar Clustering at 25 kpc from a Complete Sample of
Binaries, ApJ, submitted
Hennawi et al. 2006, Binary Quasars in the Sloan Digital Sky Survey:
Evidence for Excess Clustering on Small Scales, AJ, 131, 1
Comic Magnification
Scranton et al. 2005, Detection of Cosmic Magnification with the Sloan
Digital Sky Survey, ApJ, 633, 589
Integrated Sachs-Wolfe Effect
Giannantonio et al. 2005, High redshift detection of the integrated SachsWolfe effect, PhysRevD,74f3520
Giannantonio et al. 2008, Combined analysis of the integrated Sachs-Wolfe
effect and cosmological implications, arXiv0801.4380